Intelligent course recommendation method based on machine learning

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an intelligent course recommendation method based on machine learning, and the method comprises the steps: S1, collecting and normalizing the historical data of English courses; s2, on the basis of the normalized course data, analyzing a pre-repair relationship among courses, and constructing a directed graph representing a course dependency relationship; s3, based on the constructed course dependency graph, establishing a conversion cost model; and S4, according to the course dependency graph and the conversion cost model, learning path optimization is carried out by adopting a dynamic planning method, the total learning cost under different paths is calculated, and the path with the minimum cost is selected as a recommended learning path. According to the method, a mixed and adaptive recommendation model technical scheme is adopted, the effect of intelligently adjusting the learning path in real time is achieved, and compared with a fixed rule recommendation scheme in the prior art, the problems of student learning state feedback lagging and path planning stiffness are solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent course recommendation method based on machine learning. Background Art

[0002] In the field of English learning, course recommendation systems can help students choose courses that suit their level and learning goals. However, existing recommendation methods still have problems, which affect the learning experience and efficiency.

[0003] Most current course recommendation methods rely on fixed rules or static data. They are usually matched based on the preset course difficulty and student historical scores, and cannot adapt to students' dynamically changing learning needs. For example, some students make rapid progress in oral skills but slow improvement in reading skills. Fixed recommendation methods cannot make timely adjustments, resulting in a mismatch between course arrangements and actual needs, affecting learning outcomes.

[0004] In addition, existing technologies are still relatively crude in course structure analysis. The knowledge dependencies between courses are often organized only in a linear or hierarchical manner, ignoring the multi-level associations between knowledge points. This makes it impossible for the recommendation system to accurately identify the implicit relationships between courses, resulting in a lack of coherence in the recommendation results. Students will encounter gaps in understanding during the learning process and cannot build a complete knowledge system.

[0005] Many recommendation systems use static path planning, determine a fixed learning path in the initial stage, and recommend courses in a preset order. However, they fail to consider students' actual learning progress and real-time feedback, causing some students to enter certain courses early or late at certain stages, affecting their learning rhythm. Especially for English subjects, different students' listening, speaking, reading, and writing abilities develop unevenly, and fixed paths cannot accurately match individual needs, which in turn affects learning outcomes.

[0006] To solve the above problems, the present invention proposes an intelligent course recommendation method based on machine learning, which combines hybrid and adaptive recommendation models, English subject knowledge graphs and personalized learning path planning to realize intelligent and precise English course recommendation solutions.

[0007] Therefore, those skilled in the art provide an intelligent course recommendation method based on machine learning to solve the above-mentioned problems. Summary of the invention

[0008] In view of the deficiencies in the prior art, the present invention provides an intelligent course recommendation method based on machine learning to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent course recommendation method based on machine learning, comprising: Step S1, collecting and normalizing historical data of English courses; Step S2, based on the normalized course data, analyzing the prerequisite relationship between courses and constructing a directed graph representing the course dependency relationship; Step S3, building a conversion cost model based on the constructed course dependency graph; Step S4, based on the course dependency graph and the conversion cost model, a dynamic programming method is used to optimize the learning path, the total learning cost under different paths is calculated, and the minimum cost path is selected as the recommended learning path; Step S5, based on the optimal learning path, a reinforcement learning mechanism is introduced to dynamically adjust the path according to the learner's real-time learning progress, feedback and learning effect; Step S6, based on the learning path optimized by reinforcement learning, construct a flow network model to optimize the knowledge transfer process; Step S7: periodically update the system parameters based on the optimization results of the flow network model.

[0010] Preferably, the step S1 further comprises: Step 1.1, collect historical assessment data, course outline data and teaching feedback data of English subjects, and preliminarily organize the data; Step 1.2: Clean the collected data and remove outliers. Remove outliers by setting thresholds and complete missing data. The judgment conditions for outlier removal are set as follows: , in, is the data set after removing outliers, is the original data set, is the data point, is the data mean, is the data standard deviation; Step 1.3, use the normalization method to convert the course difficulty, learning time and expected benefits into values ​​between 0 and 1. The normalization method uses the min-max normalization method to map the data to a unified interval. The normalization calculation formula is as follows: , in, is the normalized value, is the original data, and are the minimum and maximum values ​​of the data. The normalized data will be used to construct the course dependency graph and as the input of the conversion cost model.

[0011] Preferably, the step S2 further comprises: Step 2.1: Construct a directed acyclic graph to represent the prerequisite relationship between English courses. Construct a directed acyclic graph to represent the prerequisite relationship between different English courses. There are N courses, where each course is denoted as , the dependencies between courses are represented as directed edges in the graph, and each edge Indicates the course Yes Course The prerequisite courses are as follows: , in, Represents a set of courses, Represents the set of dependency edges between courses; Step 2.2, assign normalized difficulty, learning time and expected benefit attributes to each course node: Assign the attributes of normalized difficulty, learning time and expected return, and set the attributes as follows , and , calculated by the following formula: Normalized Difficulty The complexity of the course content is scored and normalized to be expressed as , the formula is as follows: , in, For Courses The original difficulty value of This is the highest difficulty value among all courses; Study time According to the content volume and learning plan evaluation, after normalization, it is expressed as , the formula is as follows: , in, For Courses The original learning time, The longest study time among all courses; Expected Returns Represents the value of the knowledge and skills acquired by students in learning the course, which is expressed as , the formula is as follows: , in, For Courses The original expected return, The largest expected benefit of all courses; Step 2.3, divide the course dependency into layers to express cross-layer course dependency: The course dependency is divided into multiple layers, and the course layer is , indicating the course The layer you are in, assuming that the dependencies between the layers are: , in, Indicates the course Yes Course Prerequisite courses, Indicates the course Yes Course Prerequisite courses, Indicates the course The level at which you are located.

[0012] Preferably, the step S3 further comprises: Step 3.1, establish the course conversion cost function and use the weighted summation method to calculate the normalized difficulty difference, normalized learning time difference and normalized expected benefit difference between the previous and next courses. The conversion cost function formula is defined as: ; in, is the course switching cost, is the difficulty difference weight coefficient, is the learning time difference weight coefficient, is the expected return difference weight coefficient, is the normalized difficulty difference, is the normalized learning time difference between the previous and next courses, is the normalized expected return difference between the previous and next courses; Step 3.2: Determine the calculation method of the normalized difficulty difference. The normalized difficulty assigned to the previous and next courses in step S2 is expressed in the following order: and Indicates that the calculation formula is: , in, is the normalized difficulty of the previous course, is the normalized difficulty of the latter course; Step 3.3, set the fixed weight coefficient of learning time and expected return, refer to the normalized learning time and normalized expected return assigned by the previous and next courses in step S2, and use the symbol and as well as and Indicates that the calculation formula of learning time difference and expected benefit difference is defined as: , , in, , and All represent pre-set constants and satisfy .

[0013] Preferably, the step S4 further comprises: Step 4.1: Use dynamic programming to calculate the minimum cumulative conversion cost from the starting course node to the ending course node. Set the minimum cumulative conversion cost of each course node as , the minimum conversion cost of each subsequent node is calculated recursively through dynamic programming. According to the Bellman optimality principle, the minimum cumulative conversion cost of the course node is calculated by the following recursive formula: , in, For Courses The minimum cumulative conversion cost is For the previous course The minimum cumulative conversion cost is For Courses To the course The conversion cost, represents the path with the minimum switching cost; Step 4.2, use the topological sorting algorithm to determine the path with the minimum cumulative conversion cost from the starting point to the end point. The topological sorting will sort the courses according to their dependencies. Set For Courses The position in the topological sort is calculated using the following formula: , in, For Courses The position in the topological sort, is the directed acyclic graph constructed in step 2.1 Perform topological sorting; The course node sequence obtained by topological sorting is combined with the minimum cumulative conversion cost calculated in step 4.1 to determine the minimum cost path.

[0014] Preferably, the step S5 further comprises: Step 5.1, student learning status is modeled as a state in a Markov decision process, setting Indicates that students are in the course The learning status on the course includes the progress of the course and the current learning situation. The status is updated according to the student's learning progress and status changes in the course. The formula is as follows: , in, Indicates that students are in the course The state of Indicates that students are in the course On learning progress, Indicates the current course being studied; Step 5.2, course selection The actions are modeled as a Markov decision process, setting is the action selected by the student during the learning process, indicating the next course the student decides to choose. The formula is as follows: , in, represents the action selected by the student, Indicates that students are in the course The state of the upper Step 5.3: Use Q-learning algorithm to update the action value function , which is used to evaluate the expected return of taking an action in a certain state. The Q-learning algorithm is updated based on immediate feedback and future reward discount factors. For the current state Next Select Action The action value of is updated as follows: , in, Indicates in status Take action The value of learning rate, For instant rewards, is the discount factor, Indicates status Take action The maximum action value of Step 5.4, select the next course with the minimum cumulative transition cost based on the updated action value, and the student will Pick your next course , so that the cumulative conversion cost is minimized, the formula is: , in, Indicates the best course selected. Indicates in status Take action value.

[0015] Preferably, the step S6 further comprises: Step 6.1, convert the course dependency graph constructed in step 2.1 into a flow network. In the flow network, each course node represents a course, and the capacity of each edge represents the amount of knowledge transfer. The representation of the flow network is defined by the following formula: , in, is the set of nodes in the flow network, is the set of edges in the flow network, Yes Course To the course The edge capacity of For the flow network; Step 6.2, for each edge in the flow network Set the capacity limit to represent the maximum capacity of knowledge transfer between courses, and set the capacity limit of each edge to ,course To the course The knowledge transfer capacity of each capacity value It should be set based on the expected benefits of the course and the delivery capacity of the course content. The formula is as follows: , in, Yes Course To the course The capacity, Yes Course The expected return, For Courses To the course The knowledge transfer coefficient of Step 6.3, apply the maximum flow algorithm to calculate the maximum knowledge flow from the source node to the sink node, optimize the global knowledge transfer effect, set the source node to S, the sink node to T, the maximum flow to F, the maximum flow algorithm calculates the maximum flow from the source node S to the sink node T, the calculation formula of the maximum flow algorithm is: , Where F is the maximum knowledge flow from source node S to sink node T, By course To the course The actual flow rate, is the set of edges in the flow network.

[0016] Preferably, the step S7 further comprises: Step 7.1, set the system parameter update cycle symbol Indicates that system parameters are passed The group feedback cycle is updated once, and the duration of each group feedback cycle is Unit time, update cycle calculation formula is: , in, Indicates the system parameter update cycle, represents the number of feedback cycles, Indicates the length of a single group feedback cycle; Step 7.2, according to the original values ​​of the course conversion cost function coefficients obtained in step S3, they are recorded as , , The weight coefficients corresponding to the normalized difficulty difference, normalized learning time difference and normalized expected return difference; Assume that the student feedback factor is , and , the feedback update rate is recorded as , and The update formula is: , , , in, , , represents the updated course conversion cost function coefficient, , , represents the original course conversion cost function coefficient, , , represents the feedback update rate, , and It represents the real-time feedback factor for students with normalized difficulty difference, learning time difference and expected benefit difference; According to the original value of the reinforcement learning parameter obtained in step S5, it is recorded as and The learning rate and future reward discount factor in the corresponding action value function update; Assume that the student feedback factors are and , the feedback update rate is recorded as and The update formula is: , , in, , represents the updated reinforcement learning parameters, , represents the original reinforcement learning parameters, , represents the feedback update rate, and represents the student feedback factor for the learning rate and future reward discount factor; According to the original capacity of the edges between courses in the flow network obtained in step S6, it is recorded as Indicates the course To the course knowledge transfer capabilities; Let the student feedback factor be , the edge capacity update rate is recorded as The update formula is: , in, represents the updated flow network edge capacity, represents the edge capacity of the original flow network, represents the edge capacity feedback update rate, Indicates the course To the course Student feedback factor for knowledge transfer; Step 7.3, according to steps S4, S5 and S6, the updated minimum cumulative conversion cost is obtained and recorded as , the updated action value is recorded as The updated maximum knowledge flow is recorded as ; Set the aggregate recommendation score symbol It represents the comprehensive score of the next optimal course selection, and the aggregation weight coefficient is recorded as , and And meet The calculation formula is: , in, represents the aggregated recommendation score, represents the minimum cumulative conversion cost weight coefficient, represents the action value weight coefficient, represents the maximum knowledge flow weight coefficient, represents the updated minimum cumulative conversion cost obtained in step S4, represents the updated action value obtained in step S5, represents the updated maximum knowledge flow obtained in step S6.

[0017] A terminal device includes a processor and a memory, wherein the processor is used to execute an intelligent course recommendation method based on machine learning, and the memory is used to store course data and student history data related to the English subject. The terminal device can collect data, build a course dependency graph, calculate the cost of a learning path, solve the optimal learning path, perform reinforcement learning updates, and optimize global knowledge transfer.

[0018] A storage medium stores a computer program, which executes an intelligent course recommendation method based on machine learning on a terminal device. The computer program includes a data acquisition and preprocessing module, a course dependency graph construction module, a learning path cost model construction module, a dynamic programming solution module, a reinforcement learning dynamic adjustment module and a flow network global optimization module.

[0019] The present invention provides an intelligent course recommendation method based on machine learning. It has the following beneficial effects: 1. The present invention adopts a hybrid and adaptive recommendation model technical solution to achieve the effect of intelligent real-time adjustment of the learning path. Compared with the fixed rule recommendation solution in the prior art, it solves the problems of delayed feedback on students' learning status and rigid path planning.

[0020] 2. The present invention adopts the technical solution of constructing English subject knowledge graph to achieve the effect of comprehensively integrating course dependency and knowledge structure. Compared with the solution of course information fragmentation in the prior art, it solves the problems of fuzzy course dependency and discontinuous knowledge transfer.

[0021] 3. The present invention adopts a personalized learning path planning technical solution to achieve the effect of dynamically optimizing the learning process based on historical data and instant feedback. Compared with the fixed path design solution in the prior art, it solves the problems of insufficient matching of individual needs and unreasonable allocation of learning resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0023] In order to make the technical personnel in the technical field understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in combination with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a partial embodiment of the present invention, not a complete embodiment. Based on the embodiment of the present invention, other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0024] The present invention is described in detail below in conjunction with the accompanying drawings: Example

[0025] Please see attached Figure 1 , an embodiment of the present invention provides an intelligent course recommendation method based on machine learning, comprising: Step S1, collecting and normalizing historical data of English courses; Step 1.1, collect historical assessment data, course outline data and teaching feedback data of English subjects, and preliminarily organize the data; Step 1.2: Clean the collected data and remove outliers. Remove outliers by setting thresholds and complete missing data. The judgment conditions for outlier removal are set as follows: , in, is the data set after removing outliers, is the original data set, is the data point, is the data mean, is the data standard deviation; Step 1.3, use the normalization method to convert the course difficulty, learning time and expected benefits into values ​​between 0 and 1. The normalization method uses the min-max normalization method to map the data to a unified interval. The normalization calculation formula is as follows: , in, is the normalized value, is the original data, and are the minimum and maximum values ​​of the data. The normalized data will be used to construct the course dependency graph and as the input of the conversion cost model; Step S2, based on the normalized course data, analyzing the prerequisite relationship between courses and constructing a directed graph representing the course dependency relationship; Step 2.1: Construct a directed acyclic graph to represent the prerequisite relationship between English courses. Construct a directed acyclic graph to represent the prerequisite relationship between different English courses. There are N courses, where each course is denoted as , the dependencies between courses are represented as directed edges in the graph, and each edge Indicates the course Yes Course The prerequisite courses are as follows: , in, Represents a set of courses, Represents the set of dependency edges between courses; Step 2.2, assign normalized difficulty, learning time and expected benefit attributes to each course node: Assign the attributes of normalized difficulty, learning time and expected return, and set the attributes as follows , and , calculated by the following formula: Normalized Difficulty The complexity of the course content is scored and normalized to be expressed as , the formula is as follows: , in, For Courses The original difficulty value of This is the highest difficulty value among all courses; Study time According to the content volume and learning plan evaluation, after normalization, it is expressed as , the formula is as follows: , in, For Courses The original learning time, The longest study time among all courses; Expected Returns Represents the value of the knowledge and skills acquired by students in learning the course, which is expressed as , the formula is as follows: , in, For Courses The original expected return, The largest expected benefit of all courses; Step 2.3, divide the course dependency into layers to express cross-layer course dependency: The course dependency is divided into multiple layers, and the course layer is , indicating the course The layer you are in, assuming that the dependencies between the layers are: , in, Indicates the course Yes Course Prerequisite courses, Indicates the course Yes Course Prerequisite courses, Indicates the course The level at which you are located; Step S3, building a conversion cost model based on the constructed course dependency graph; Step 3.1, establish the course conversion cost function and use the weighted summation method to calculate the normalized difficulty difference, normalized learning time difference and normalized expected benefit difference between the previous and next courses. The conversion cost function formula is defined as: ; in, is the course switching cost, is the difficulty difference weight coefficient, is the learning time difference weight coefficient, is the expected return difference weight coefficient, is the normalized difficulty difference, is the normalized learning time difference between the previous and next courses, is the normalized expected return difference between the previous and next courses; Step 3.2: Determine the calculation method of the normalized difficulty difference. The normalized difficulty assigned to the previous and next courses in step S2 is expressed in the following order: and Indicates that the calculation formula is: , in, is the normalized difficulty of the previous course, is the normalized difficulty of the latter course; Step 3.3, set the fixed weight coefficient of learning time and expected return, refer to the normalized learning time and normalized expected return assigned by the previous and next courses in step S2, and use the symbol and as well as and Indicates that the calculation formula of learning time difference and expected benefit difference is defined as: , , in, , and All represent pre-set constants and satisfy .

[0026] Step S4, based on the course dependency graph and the conversion cost model, a dynamic programming method is used to optimize the learning path, the total learning cost under different paths is calculated, and the minimum cost path is selected as the recommended learning path; Step 4.1: Use dynamic programming to calculate the minimum cumulative conversion cost from the starting course node to the ending course node. Set the minimum cumulative conversion cost of each course node as , the minimum conversion cost of each subsequent node is calculated recursively through dynamic programming. According to the Bellman optimality principle, the minimum cumulative conversion cost of the course node is calculated by the following recursive formula: , in, For Courses The minimum cumulative conversion cost is For the previous course The minimum cumulative conversion cost is For Courses To the course The conversion cost, represents the path with the minimum switching cost; Step 4.2, use the topological sorting algorithm to determine the path with the minimum cumulative conversion cost from the starting point to the end point. The topological sorting will sort the courses according to their dependencies. Set For Courses The position in the topological sort is calculated using the following formula: , in, For Courses The position in the topological sort, is the directed acyclic graph constructed in step 2.1 Perform topological sorting; The course node sequence obtained by topological sorting is combined with the minimum cumulative conversion cost calculated in step 4.1 to determine the minimum cost path.

[0027] Step S5, based on the optimal learning path, a reinforcement learning mechanism is introduced to dynamically adjust the path according to the learner's real-time learning progress, feedback and learning effect; Step 5.1, student learning status is modeled as a state in a Markov decision process, setting Indicates that students are in the course The learning status on the course includes the progress of the course and the current learning situation. The status is updated according to the student's learning progress and status changes in the course. The formula is as follows: , in, Indicates that students are in the course The state of Indicates that students are in the course On learning progress, Indicates the current course being studied; Step 5.2, course selection The actions are modeled as a Markov decision process, setting is the action selected by the student during the learning process, indicating the next course the student decides to choose. The formula is as follows: , in, represents the action selected by the student, Indicates that students are in the course The state of the upper Step 5.3: Use Q-learning algorithm to update the action value function , which is used to evaluate the expected return of taking an action in a certain state. The Q-learning algorithm is updated based on immediate feedback and future reward discount factors. For the current state Next Select Action The action value of is updated as follows: , in, Indicates in status Take action The value of learning rate, For instant rewards, is the discount factor, Indicates status Take action The maximum action value of Step 5.4, select the next course with the minimum cumulative transition cost based on the updated action value, and the student will Pick your next course , so that the cumulative conversion cost is minimized, the formula is: , in, Indicates the best course selected. Indicates in status Take action the value of Step S6, based on the learning path optimized by reinforcement learning, construct a flow network model to optimize the knowledge transfer process; Step 6.1, convert the course dependency graph constructed in step 2.1 into a flow network. In the flow network, each course node represents a course, and the capacity of each edge represents the amount of knowledge transfer. The representation of the flow network is defined by the following formula: , in, is the set of nodes in the flow network, is the set of edges in the flow network, Yes Course To the course The edge capacity of For the flow network; Step 6.2, for each edge in the flow network Set the capacity limit to represent the maximum capacity of knowledge transfer between courses, and set the capacity limit of each edge to ,course To the course The knowledge transfer capacity of each capacity value It should be set based on the expected benefits of the course and the delivery capacity of the course content. The formula is as follows: , in, Yes Course To the course The capacity, Yes Course The expected return, For Courses To the course The knowledge transfer coefficient of Step 6.3, apply the maximum flow algorithm to calculate the maximum knowledge flow from the source node to the sink node, optimize the global knowledge transfer effect, set the source node to S, the sink node to T, the maximum flow to F, the maximum flow algorithm calculates the maximum flow from the source node S to the sink node T, the calculation formula of the maximum flow algorithm is: , Where F is the maximum knowledge flow from source node S to sink node T, By course To the course The actual flow rate, is the set of edges in the flow network.

[0028] Step S7, periodically updating system parameters in combination with the optimization results of the flow network model; Step 7.1, set the system parameter update cycle symbol Indicates that system parameters are passed The group feedback cycle is updated once, and the duration of each group feedback cycle is Unit time, update cycle calculation formula is: , in, Indicates the system parameter update cycle, represents the number of feedback cycles, Indicates the length of a single group feedback cycle; Step 7.2, according to the original values ​​of the course conversion cost function coefficients obtained in step S3, they are recorded as , , The weight coefficients corresponding to the normalized difficulty difference, normalized learning time difference and normalized expected return difference; Assume that the student feedback factor is , and , the feedback update rate is recorded as , and The update formula is: , , , in, , , represents the updated course conversion cost function coefficient, , , represents the original course conversion cost function coefficient, , , represents the feedback update rate, , and It represents the real-time feedback factor for students with normalized difficulty difference, learning time difference and expected benefit difference; According to the original value of the reinforcement learning parameter obtained in step S5, it is recorded as and The learning rate and future reward discount factor in the corresponding action value function update; Assume that the student feedback factors are and , the feedback update rate is recorded as and The update formula is: , , in, , represents the updated reinforcement learning parameters, , represents the original reinforcement learning parameters, , represents the feedback update rate, and represents the student feedback factor for the learning rate and future reward discount factor; According to the original capacity of the edges between courses in the flow network obtained in step S6, it is recorded as Indicates the course To the course knowledge transfer capabilities; Let the student feedback factor be , the edge capacity update rate is recorded as The update formula is: , in, represents the updated flow network edge capacity, represents the edge capacity of the original flow network, represents the edge capacity feedback update rate, Indicates the course To the course Student feedback factor for knowledge transfer; Step 7.3, according to steps S4, S5 and S6, the updated minimum cumulative conversion cost is obtained and recorded as , the updated action value is recorded as The updated maximum knowledge flow is recorded as ; Set the aggregate recommendation score symbol It represents the comprehensive score of the next optimal course selection, and the aggregation weight coefficient is recorded as , and And meet The calculation formula is: , in, represents the aggregated recommendation score, represents the minimum cumulative conversion cost weight coefficient, represents the action value weight coefficient, represents the maximum knowledge flow weight coefficient, represents the updated minimum cumulative conversion cost obtained in step S4, represents the updated action value obtained in step S5, represents the updated maximum knowledge flow obtained in step S6.

[0029] Data normalization and exception handling (S1): Normalized data allows course information of different dimensions to be compared on the same basis, preventing certain features from affecting recommendations due to scale differences. Outlier removal is like "filtering junk data" to ensure that reliable information is input and to avoid incorrect data affecting the learning path.

[0030] Course dependency analysis (S2), directed acyclic graph can accurately express the selection of basic courses and advanced courses, preventing students from skipping key knowledge points and confusing the learning order.

[0031] Switching cost model (S3): During the learning process, different courses have different difficulty levels and different learning times. The switching cost model can calculate the "cost" of switching courses to ensure that the recommended learning path is reasonable and avoid jumping from a particularly easy course to a difficult course.

[0032] Dynamic path optimization (S4): Traditional learning paths are fixed at the beginning, but each person's learning situation changes dynamically. The present invention adopts a dynamic programming algorithm, and each recommendation will comprehensively consider the current learning progress, calculate the optimal path, reduce learning costs, and ensure learning efficiency.

[0033] Reinforcement learning mechanism (S5): Students’ learning habits and status are changing, and rigid recommendation schemes cannot adapt. This method introduces the Q-learning reinforcement learning algorithm, which is equivalent to an “intelligent regulator”. It will continuously optimize the learning path based on students’ real-time feedback to ensure that the recommended content is always suitable for the current level.

[0034] The flow network optimizes knowledge transfer (S6). The knowledge points between courses flow like "information flow", and a course can provide key knowledge for subsequent courses. The present invention adopts the maximum flow algorithm to ensure the smoothness of knowledge transfer. It does not recommend courses, but calculates how to maximize the absorption of knowledge and improve the overall learning effect.

[0035] System parameters are updated periodically (S7). Traditional recommendation systems are often "fixed" and cannot change with the progress of students. This system will periodically adjust the recommendation parameters, and can continuously optimize and make accurate recommendations based on students' actual feedback to meet individual needs.

[0036] In summary, the intelligent course recommendation system is not simply "pushing courses", but combining machine learning, reinforcement learning and graph algorithms to make the learning path scientific, the recommendation intelligent, and the knowledge absorption efficient. Compared with the traditional fixed rule recommendation, this method can adjust the learning path in real time, dynamically optimize the recommended content, and intelligently match personalized needs to ensure that each student can get the most suitable English learning plan.

[0037] A terminal device includes a processor and a memory, wherein the processor is used to execute an intelligent course recommendation method based on machine learning, and the memory is used to store course data and student history data related to the English subject. The terminal device can collect data, build a course dependency graph, calculate the cost of a learning path, solve the optimal learning path, execute reinforcement learning updates, and optimize global knowledge transfer.

[0038] A storage medium stores a computer program, which executes an intelligent course recommendation method based on machine learning on a terminal device. The computer program includes a data acquisition and preprocessing module, a course dependency graph construction module, a learning path cost model construction module, a dynamic programming solution module, a reinforcement learning dynamic adjustment module and a flow network global optimization module.

[0039] The traditional course recommendation method is rigid, and students progress at different speeds, but the recommendation system is the same. This method uses a hybrid and adaptive recommendation model to allow the learning path to be adjusted in real time, following the student's learning pace, and avoiding the disconnection between learning progress and recommended content.

[0040] The relationship between courses is complex, and many traditional systems simply recommend courses by difficulty, resulting in loose connections between knowledge points. This invention uses the English subject knowledge graph to accurately identify course dependencies, ensuring that the recommended learning path is scientific and efficient, and avoiding students from "breaking the learning process" and missing key knowledge.

[0041] Each student has different foundations and learning styles, but the existing system often only makes recommendations based on historical performance and cannot be adjusted dynamically. This method uses personalized learning path planning, combined with students' historical data, real-time feedback and learning effects, to dynamically optimize the learning path so that all individuals can get the most suitable course arrangement for themselves.

[0042] This invention breaks through the limitations of traditional English course recommendations and combines machine learning, reinforcement learning, knowledge graphs and dynamic programming to create an intelligent and accurate personalized learning recommendation system.

[0043] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent course recommendation method based on machine learning, characterized in that: include: Step S1, collecting and normalizing historical data of English courses; Step S2, based on the normalized course data, analyzing the prerequisite relationship between courses and constructing a directed graph representing the course dependency relationship; Step S3, building a conversion cost model based on the constructed course dependency graph; Step S4, based on the course dependency graph and the conversion cost model, a dynamic programming method is used to optimize the learning path, the total learning cost under different paths is calculated, and the minimum cost path is selected as the recommended learning path; Step S5, based on the optimal learning path, a reinforcement learning mechanism is introduced to dynamically adjust the path according to the learner's real-time learning progress, feedback and learning effect; Step S6, based on the learning path optimized by reinforcement learning, construct a flow network model to optimize the knowledge transfer process; Step S7: periodically update the system parameters based on the optimization results of the flow network model.

2. The intelligent course recommendation method based on machine learning according to claim 1, characterized in that: The step S1 further comprises: Step 1.1, collect historical assessment data, course outline data and teaching feedback data of English subjects, and preliminarily organize the data; Step 1.2: Clean the collected data and remove outliers. Remove outliers by setting thresholds and complete missing data. The judgment conditions for outlier removal are set as follows: ， in, is the data set after removing outliers, is the original data set, is the data point, is the data mean, is the data standard deviation; Step 1.3, use the normalization method to convert the course difficulty, learning time and expected benefits into values ​​between 0 and 1. The normalization method uses the min-max normalization method to map the data to a unified interval. The normalization calculation formula is as follows: , in, is the normalized value, is the original data, and are the minimum and maximum values ​​of the data. The normalized data will be used to construct the course dependency graph and as the input of the conversion cost model.

3. The intelligent course recommendation method based on machine learning according to claim 1, characterized in that: The step S2 further comprises: Step 2.1: Construct a directed acyclic graph to represent the prerequisite relationship between English courses. Construct a directed acyclic graph to represent the prerequisite relationship between different English courses. There are N courses, where each course is denoted as , the dependencies between courses are represented as directed edges in the graph, and each edge Indicates the course Yes Course The prerequisite courses are as follows: ， in, Represents a set of courses, Represents the set of dependency edges between courses; Step 2.2, assign normalized difficulty, learning time and expected benefit attributes to each course node: Assign the attributes of normalized difficulty, learning time and expected return, and set the attributes as follows , and , calculated by the following formula: Normalized Difficulty The complexity of the course content is scored and normalized to be expressed as , the formula is as follows: , in, For Courses The original difficulty value of This is the highest difficulty value among all courses; Study time According to the content volume and learning plan evaluation, after normalization, it is expressed as , the formula is as follows: , in, For Courses The original learning time, The longest study time among all courses; Expected Returns Represents the value of the knowledge and skills acquired by students in learning the course, which is expressed as , the formula is as follows: , in, For Courses The original expected return, The largest expected benefit of all courses; Step 2.3, divide the course dependency into layers to express cross-layer course dependency: The course dependency is divided into multiple layers, and the course layer is , indicating the course The layer you are in, assuming that the dependencies between the layers are: ， in, Indicates the course Yes Course Prerequisite courses, Indicates the course Yes Course Prerequisite courses, Indicates the course The level at which you are located.

4. The intelligent course recommendation method based on machine learning according to claim 1, characterized in that: The step S3 further comprises: Step 3.1, establish the course conversion cost function and use the weighted summation method to calculate the normalized difficulty difference, normalized learning time difference and normalized expected benefit difference between the previous and next courses. The conversion cost function formula is defined as: ; in, is the course switching cost, is the difficulty difference weight coefficient, is the learning time difference weight coefficient, is the expected return difference weight coefficient, is the normalized difficulty difference, is the normalized learning time difference between the previous and next courses, is the normalized expected return difference between the previous and next courses; Step 3.2: Determine the calculation method of the normalized difficulty difference. The normalized difficulty assigned to the previous and next courses in step S2 is expressed in the following order: and Indicates that the calculation formula is: ， in, is the normalized difficulty of the previous course, is the normalized difficulty of the latter course; Step 3.3, set the fixed weight coefficient of learning time and expected return, refer to the normalized learning time and normalized expected return assigned by the previous and next courses in step S2, and use the symbol and as well as and Indicates that the calculation formula of learning time difference and expected benefit difference is defined as: ， ， in, , and All represent pre-set constants and satisfy .

5. The intelligent course recommendation method based on machine learning according to claim 1, characterized in that: The step S4 further comprises: Step 4.1: Use dynamic programming to calculate the minimum cumulative conversion cost from the starting course node to the ending course node. Set the minimum cumulative conversion cost of each course node as , the minimum conversion cost of each subsequent node is calculated recursively through dynamic programming. According to the Bellman optimality principle, the minimum cumulative conversion cost of the course node is calculated by the following recursive formula: ， in, For Courses The minimum cumulative conversion cost is For the previous course The minimum cumulative conversion cost is For Courses To the course The conversion cost, represents the path with the minimum switching cost; Step 4.2, use the topological sorting algorithm to determine the path with the minimum cumulative conversion cost from the starting point to the end point. The topological sorting will sort the courses according to their dependencies. Set For Courses The position in the topological sort is calculated using the following formula: ， in, For Courses The position in the topological sort, is the directed acyclic graph constructed in step 2.1 Perform topological sorting; The course node sequence obtained by topological sorting is combined with the minimum cumulative conversion cost calculated in step 4.1 to determine the minimum cost path.

6. The intelligent course recommendation method based on machine learning according to claim 1, characterized in that: The step S5 further comprises: Step 5.1, student learning status is modeled as a state in a Markov decision process, setting Indicates that students are in the course The learning status on the course includes the progress of the course and the current learning situation. The status is updated according to the student's learning progress and status changes in the course. The formula is as follows: , in, Indicates that students are in the course The state of Indicates that students are in the course On learning progress, Indicates the current course being studied; Step 5.2, course selection The actions are modeled as a Markov decision process, setting is the action selected by the student during the learning process, indicating the next course the student decides to choose. The formula is as follows: ， in, represents the action selected by the student, Indicates that students are in the course The state of the upper Step 5.3: Use Q-learning algorithm to update the action value function , which is used to evaluate the expected return of taking an action in a certain state. The Q-learning algorithm is updated based on immediate feedback and future reward discount factors. For the current state Next Select Action The action value of is updated as follows: ， in, Indicates in status Take action The value of learning rate, For instant rewards, is the discount factor, Indicates status Take action The maximum action value of Step 5.4, select the next course with the minimum cumulative transition cost based on the updated action value, and the student will Pick your next course , so that the cumulative conversion cost is minimized, the formula is: , in, Indicates the best course selected. Indicates in status Take action value.

7. The intelligent course recommendation method based on machine learning according to claim 1, characterized in that: The step S6 further comprises: Step 6.1, convert the course dependency graph constructed in step 2.1 into a flow network. In the flow network, each course node represents a course, and the capacity of each edge represents the amount of knowledge transfer. The representation of the flow network is defined by the following formula: , in, is the set of nodes in the flow network, is the set of edges in the flow network, Yes Course To the course The edge capacity of For the flow network; Step 6.2, for each edge in the flow network Set the capacity limit to represent the maximum capacity of knowledge transfer between courses, and set the capacity limit of each edge to ,course To the course The knowledge transfer capacity of each capacity value It should be set based on the expected benefits of the course and the delivery capacity of the course content. The formula is as follows: ， in, Yes Course To the course The capacity, Yes Course The expected return, For Courses To the course The knowledge transfer coefficient of Step 6.3, apply the maximum flow algorithm to calculate the maximum knowledge flow from the source node to the sink node, optimize the global knowledge transfer effect, set the source node to S, the sink node to T, the maximum flow to F, the maximum flow algorithm calculates the maximum flow from the source node S to the sink node T, the calculation formula of the maximum flow algorithm is: ， Where F is the maximum knowledge flow from source node S to sink node T, By course To the course The actual flow rate, is the set of edges in the flow network.

8. The intelligent course recommendation method based on machine learning according to claim 1, characterized in that: The step S7 further comprises: Step 7.1, set the system parameter update cycle symbol Indicates that system parameters are passed The group feedback cycle is updated once, and the duration of each group feedback cycle is Unit time, update cycle calculation formula is: ， in, Indicates the system parameter update cycle, represents the number of feedback cycles, Indicates the length of a single group feedback cycle; Step 7.2, according to the original values ​​of the course conversion cost function coefficients obtained in step S3, they are recorded as , , The weight coefficients corresponding to the normalized difficulty difference, normalized learning time difference and normalized expected return difference; Assume that the student feedback factor is , and , the feedback update rate is recorded as , and The update formula is: ， ， ， in, , , represents the updated course conversion cost function coefficient, , , represents the original course conversion cost function coefficient, , , represents the feedback update rate, , and It represents the real-time feedback factor for students with normalized difficulty difference, learning time difference and expected benefit difference; According to the original value of the reinforcement learning parameter obtained in step S5, it is recorded as and The learning rate and future reward discount factor in the corresponding action value function update; Assume that the student feedback factors are and , the feedback update rate is recorded as and The update formula is: ， ， in, , represents the updated reinforcement learning parameters, , represents the original reinforcement learning parameters, , represents the feedback update rate, and represents the student feedback factor for the learning rate and future reward discount factor; According to the original capacity of the edges between courses in the flow network obtained in step S6, it is recorded as Indicates the course To the course knowledge transfer capabilities; Let the student feedback factor be , the edge capacity update rate is recorded as The update formula is: ， in, represents the updated flow network edge capacity, represents the edge capacity of the original flow network, represents the edge capacity feedback update rate, Indicates the course To the course Student feedback factor for knowledge transfer; Step 7.3, according to steps S4, S5 and S6, the updated minimum cumulative conversion cost is obtained and recorded as , the updated action value is recorded as The updated maximum knowledge flow is recorded as ; Set the aggregate recommendation score symbol It represents the comprehensive score of the next optimal course selection, and the aggregation weight coefficient is recorded as , and And meet The calculation formula is: ， in, represents the aggregated recommendation score, represents the minimum cumulative conversion cost weight coefficient, represents the action value weight coefficient, represents the maximum knowledge flow weight coefficient, represents the updated minimum cumulative conversion cost obtained in step S4, represents the updated action value obtained in step S5, represents the updated maximum knowledge flow obtained in step S6.

9. A terminal device, characterized in that: It includes a processor and a memory, the processor is used to execute an intelligent course recommendation method based on machine learning, the memory is used to store course data and student history data related to the English subject, and the terminal device can collect data, build a course dependency graph, calculate the cost of the learning path, solve the optimal learning path, perform reinforcement learning updates and optimize global knowledge transfer.

10. A storage medium, characterized in that: A computer program is stored, which executes an intelligent course recommendation method based on machine learning on a terminal device. The computer program includes a data acquisition and preprocessing module, a course dependency graph construction module, a learning path cost model construction module, a dynamic programming solution module, a reinforcement learning dynamic adjustment module and a flow network global optimization module.

Images